Jamie McGowan

I am a Research Scientist at MediaTek Research. My research is focussed on the training dynamics of neural networks and the interpretibility of their predictions.

My (vague and ambitious) research aim is to develop a deeper knowledge of how and why deep neural networks learn in specific ways and use this understanding to inform the next generation of AI. In particular, I’m interested in using this knowledge to design simpler algorithms and architectures, especially when applied to multimodality.

Some very broad questions I have been thinking about recently include:

• How can neural networks become plastic?
  • Plasticity is related to how well a neural network can adapt to new domains and/or modalities.
• How can task-dependent behaviour be extracted in deep networks?
  • Or how can we insert this?
• What is next after Adam?
  • Or are we going to be using it forever and ever…

My background and training is in Theoretical Physics where I worked on extending our understanding of the Standard Model in a data-driven environment.

If you are interested in collaboration or our work/opportunities at MediaTek Research, please contact me here.